webml-polyfill
webml-polyfill copied to clipboard
Add the OpenVINO.js framework
The OpenVINO.js framework exists in the example for image classification, semantic_segmentation, speech_commands, and speech_recognition. It works when the Electron Application with integrated node.js.
@BruceDai @ibelem Here is the result of the comparison between the WebNN on the Chrome and OpenVINO.js on Electron
Environment: Environment: Electron v11.01 Chrome: 86.0.4209.0 OpenVINO 2020.3.194 Windows 10 TGL i7-1165G7
- Image Classification Example
Model | WebNN Perf(FAST_SINGLE_ANSWER)(ms) | OpenVINO.js Perf(ms) |
---|---|---|
SqueezeNet | 3.14 | 3.68 |
MobileNet v1 | 4.58 | 4.62 |
MobileNet v2 | 3.89 | 4.10 |
ResNet50 v1 | 19.6 | 20.7 |
DenseNet 121 | 15.58 | 15.43 |
Inception v2 | 10.68 | 12.87 |
Inception v4 | 57.54 | 57.51 |
- Semantic Segmentation Example
Model | WebNN Perf(FAST_SINGLE_ANSWER)(ms) | OpenVINO.js Perf(ms) |
---|---|---|
Deeplab_224_Atrous | 12.58 | 13 |
Deeplab_224_Quant | 7.27 | 7.55 |
Deeplab_257_Atrous | 15.13 | 16.7 |
Deeplab_257_Quant | 8.07 | 8.37 |
Deeplab_321_Atrous | 22.47 | 23.82 |
Deeplab_321_Quant | 10.84 | 11.93 |
Deeplab_513_Atrous | 59.99 | 61.27 |
Deeplab_513_Quant | 25.72 | 28.38 |
- Speech Commands Example
Model | WebNN Perf(FAST_SINGLE_ANSWER)(ms) | OpenVINO.js Perf(ms) |
---|---|---|
KWS DNN (Yes Example) | 0.43 | 0.28 |
- Speech Recognition Example
Model | WebNN Perf(FAST_SINGLE_ANSWER)(ms) | OpenVINO.js Perf(ms) |
---|---|---|
wsj_dnn5b | 2.48 | 2.37 |